Is it a full clean-room implementation? I really want to see how the AI clean-room reverse engineering projects fare against legal challenges.
Supposedly clean-room implementations are legal. But they're so easy with AI - it seems like that will need to change. Or the world will come crashing down. Hard to say which is more likely right now.
Virus Total:
Activity related to AMADEY - according to source Cluster25 - 6 months ago
This DOMAIN is used by AMADEY. Amadey is a botnet agent that appeared around October 2018 sold on Russian-speaking hacking forums. It periodically sends information about the system and installed AV software to its C2 server. It can also load other payloads to the victims' devices.
If you copy the url at the top and resolve it, it downloads the bin. But i think this may be a false alarm and the botnet just spread his malware over github.
I love trying out new software, but with Github getting blasted all the time, supply chain compromise, and now in the vibe coding era with people not even checking, we have to be careful. It sucks that this is the status quo now.
What advantages does this offer over Darktable, the long standing FOSS lightroom alternative?
Edit: One criticism of Darktable is that it emphasizes its "scene-referred workflow" as opposed to a "display-referred workflow" like Lightroom does, making it not a true Lightroom alternative. Seems like this also happens to be written in C++ and also uses a scene-referred workflow? Quite the coincidence. How much of the important design decisions (or implementation) are just copied from Darktable?
Fair question. The scene-referred design is modelled on darktable’s, and darktable deserves the credit for it. My very first prototype ported some of darktable’s algorithms, and I replaced that with an independent implementation before releasing. Both projects are GPL-3.0, and the README now credits darktable directly.
I built Rembrandt because darktable’s learning curve kept me from using it, not because it’s worse; it has far more control. The goal here is approachability: fewer, broader controls, on-device AI for masks and denoise, and Lightroom catalog import.
Stack: the engine is JavaScript with WebGL/WebGPU shaders, in a small Rust (Tauri) desktop shell. LibRaw (C++) decodes RAW files.
Edit: corrected my earlier reply. I wrongly said it was written in C++, and I overstated the darktable answer.
The only surprising thing in your comment is that the day's latest vibe coded app actually got upvoted.
I mean really, you had to ask Claude whether the workflow was Darktable-inspired. The people in this thread who read Claude's README probably know more about how "your" software works than you do.
The scene-referred workflow is vastly superior because it makes all the adjustments before the full Raw data is converted to the display space.
You don't even need to know what that means for it to be useful. In practice, it just means less artifacts, less weird stuff that happens with color near clipping highlights, etc. You use the software in exactly the same way you'd use the display-referred workflow, except that under the hood, the modules are working with the linear Raw data.
This is unfortunately exactly the kind of project I wouldn't want to trust my photo management workflow to...software is cheaper than ever, thoughtful maintenance and longevity is worth much more to me.
AI means that anyone at all can pay tiny amounts of money for software that fits their exact needs.
This excites them and they share it under the assumption someone else has their exact needs but lacks the inclination to have AI make one of their own.
The real question is why does HN keep upvoting them?
There have been at least 10 (probably more) photo editing suites. It seems that now everyone who doesn't want to learn the existing suites just can make their own with AI.
Pro photographer here. And I've counted at least ten of these photo editing apps that have appeared in the last year, many of them in the last few months. All of them LLM-assisted or completely vibe coded.
They definitely work but...
Unlike Lightroom or Darktable (an open source Raw editor that took years to get to a sophisticated level), they are all rough around the edges. Maybe in the future they'll get better in terms of being really refined to handle advanced photography, but right now they're hobby applications.
To me this highlights the absurdity of AI, because enormous amounts of energy is wasted on duplicating stuff and making something new.
I won't try and argue that nothing nice can come out of it, but it seems absurd to me to waste so much energy as a species this way. I wouldn't be surprised if there were a hundred more of these apps in another year.
It's got the kind of UI that I would love pre-2024 or so but these days it just screams "AI generic" unfortunately. Documentation is fairy concise and reasonably LLM-jargon free though.
perhaps my requirements aren’t that complex but i find Affinity mostly a great replacement for InDesign. there are some annoyances and unintuitive limitations but its all workable.
Given that some are now reversing all closed software using LLMs, it should be now finally worth reversing the whole of Nvidia’s closed source software, drivers and blobs, including CUDA.
I think they're not reversing the way you may think(and they're doing a very bad job, if you try any of these apps they collapse very soon for basic workflow things) but rather copying the behaviour by looking at input and output. A clean-room design is called iirc
There's Asahi Linux, why arent they doing it already? Imo the end result is not great and it's very hard to maintain. Like Asahi has already demonstrated
Supposedly clean-room implementations are legal. But they're so easy with AI - it seems like that will need to change. Or the world will come crashing down. Hard to say which is more likely right now.
---
Virus Total: Activity related to AMADEY - according to source Cluster25 - 6 months ago This DOMAIN is used by AMADEY. Amadey is a botnet agent that appeared around October 2018 sold on Russian-speaking hacking forums. It periodically sends information about the system and installed AV software to its C2 server. It can also load other payloads to the victims' devices.
macOS download url after downloading with librewolf and rightclicking "copy downoload url": https://www.virustotal.com/gui/url/f1e08a34d17fd85cd8896ba86...
Edit: i dont get this when copying direct links from the readme.md. Its just when i click them and copy the url it used to download.
Edit2: i get this too when i download the x64 exe and copy the download link from the librewolf (firefox fork) download manager https://www.virustotal.com/gui/url-analysis/u-6a55e124c22f9b...
Edit3: maybe false alarm and the botnet used the github url https://release-assets.githubusercontent.com/ to spread malware?
Edit: One criticism of Darktable is that it emphasizes its "scene-referred workflow" as opposed to a "display-referred workflow" like Lightroom does, making it not a true Lightroom alternative. Seems like this also happens to be written in C++ and also uses a scene-referred workflow? Quite the coincidence. How much of the important design decisions (or implementation) are just copied from Darktable?
I built Rembrandt because darktable’s learning curve kept me from using it, not because it’s worse; it has far more control. The goal here is approachability: fewer, broader controls, on-device AI for masks and denoise, and Lightroom catalog import.
Stack: the engine is JavaScript with WebGL/WebGPU shaders, in a small Rust (Tauri) desktop shell. LibRaw (C++) decodes RAW files.
Edit: corrected my earlier reply. I wrongly said it was written in C++, and I overstated the darktable answer.
I mean really, you had to ask Claude whether the workflow was Darktable-inspired. The people in this thread who read Claude's README probably know more about how "your" software works than you do.
You don't even need to know what that means for it to be useful. In practice, it just means less artifacts, less weird stuff that happens with color near clipping highlights, etc. You use the software in exactly the same way you'd use the display-referred workflow, except that under the hood, the modules are working with the linear Raw data.
Not that Adobe's getting my money either.
This compares more closely with Darktable, RawTherapee, ART etc.
This excites them and they share it under the assumption someone else has their exact needs but lacks the inclination to have AI make one of their own.
The real question is why does HN keep upvoting them?
They definitely work but...
Unlike Lightroom or Darktable (an open source Raw editor that took years to get to a sophisticated level), they are all rough around the edges. Maybe in the future they'll get better in terms of being really refined to handle advanced photography, but right now they're hobby applications.
To me this highlights the absurdity of AI, because enormous amounts of energy is wasted on duplicating stuff and making something new.
I won't try and argue that nothing nice can come out of it, but it seems absurd to me to waste so much energy as a species this way. I wouldn't be surprised if there were a hundred more of these apps in another year.
Where that's good, I don't know..nobody seems to care telling.
No excuses this time.